`utils.get_value` raise TypeError with out-of-range `int` index

[thanhlecongg]

Issue Description

When passing an out-of-range int index to utils.get_value, a TypeError is raised instead of returning the provided default value. Based on the documentation and function’s behavior in string cases, I’d expect it to fall back to default when the index is invalid. FYI, this issue seems to occur because handling IndexError requires falling back to getattr(obj, key, default). This logic was introduced in PR #1063 following git blame. I think this issue can be fixed by handling this TypeError to return default value. I can make an PR if this change make sense to you.

Reproducing Code Example

from marshmallow import utils

### Case 1: List
lst = [0, 1, 2, 3, 4, 5]
print(utils.get_value(lst, 999, default=3))

### Case 1: Dictionary with integer keys
dictionary = {1: 'a', 2: 'b', 3: 'c'}
print(utils.get_value(dictionary, 4, default='z'))

### Case 3: List of object
class PointClass:
    def __init__(self, x, y):
        self.x = x
        self.y = y

list_obj = [[PointClass(24, 42),{"x": 24, "y": 42}]]
print(utils.get_value(list_obj, 3, default=None))

Expected Result

Case 1: 3
Case 2: z
Case 3: None

Actual Behavior

Traceback (most recent call last):
  File "test.py", line 3, in <module>
    print(utils.get_value(lst, 999, default=3))
  File ".../site-packages/marshmallow/utils.py", line 108, in get_value
    return _get_value_for_key(obj, key, default)
  File ".../site-packages/marshmallow/utils.py", line 126, in _get_value_for_key
    return getattr(obj, key, default)
TypeError: getattr(): attribute name must be string

[deckar01]

get_value is only used to serialize data from the output of Schema.load(). Since enveloping the output with classes is a documented use case, it also handles class attributes. Schema.get_attribute() was designed to be overridden to provide an accessor for other types of envelopes (like lists).

We should consider breaking the attribute accessor out into a separate utility that users have to opt into. Toasted dropped attribute access as one of its primary optimizations.

We should probably also change the list test from (#340). That PR didn't add support for lists.

[deckar01]

get_value works fine for accessing integer keys on dicts. get_value has not supported lists for the last 13 years. ...

#2898 (comment)

Given this context, this issue is effectively a feature request to add list support to get_value.

The field attribute only supports strings...

https://marshmallow.readthedocs.io/en/latest/marshmallow.fields.html#marshmallow.fields.Field

Initially I was tempted to add int handling for the attribute arg, but python technically allows any hashable type to be a dict key. Since post_load must be used to envelope these types of keys, it seems reasonable to require pre_dump invert the operation. This use case is already supported, so I would advocate against adding list support to either attribute or get_value.

[thanhlecongg]

That makes sense to me. At first, I also thought get_value may be not designed to use with a list. But after going through the documentation and the related tests, I changed my mind. That being said, I think it might help to include more details about the unsupported feature, e.g., through type annotation or docstring. Or we can add a check to raise more informative value. as current TypeError seems to be confusing to me. Also, this test case should be removed: https://github.com/marshmallow-code/marshmallow/blob/dev/tests/test_utils.py#L72C1-L79C47

[lafrech]

This issue seems to be seen as an easy one for AIs to fix, so let's make things more explicit.

What the comments above mean is that current behaviour is fine. The issue is still open as a reminder that it might be nice to add a comment somewhere someday to clarify the intent. But we don't intend to change the code.

Now, new PRs duplicating the ones linked above will be sent by users that didn't care to read the issue being "fixed" using AIs that didn't either.